TL;DR
- Agentic commerce is a transaction model where AI agents complete purchases autonomously for humans, within rules the user defines upfront.
- Agentic payment execution happens in real-time: event detection → policy evaluation → routing → settlement. Policy is signed once, executed many times.
- Common business use cases include automated vendor and supplier payments, subscription and recurring payments, cross-border B2B payment automation, and agent-assisted procurement.
- Market scale is sizing up: McKinsey forecasts $3 to $5 trillion in agentic commerce sales globally by 2030. Major frameworks are already shipping: Visa Intelligent Commerce, Coinbase's x402, and more.
- Reap operates at the card-issuing layer underneath agent platforms. As one of the first licensed issuers to bring Visa Intelligent Commerce live, Reap provides agent-specific tokenized credentials, programmable spend controls, and passkey-based authentication through a single API.
Introduction
Search "Buy for Me" on a flagship retailer's app today and an AI agent will scan dozens of merchants, compare specs, find the best price, and complete the purchase, all while you finish your coffee.
In March 2026, an AI agent booked a ride to Singapore's Changi Airport in Mastercard's first live agentic transaction. In May 2026, AWS launched Amazon Bedrock AgentCore Payments, powered by Coinbase's x402 protocol, giving developers a managed path for AI agents to handle spending controls and compliance inside enterprise workflows. We have quickly evolved from utilising generative AI for recommendations to using agents to help make these purchases.
That is what users and developers are seeing today: agents shopping on consumer front ends, agents transacting inside enterprise workflows. Behind both is a question the platforms enabling them have to solve: how does an AI agent actually pay? AI-native platforms, fintechs, and the card programs underneath them are running into the same problem set. The payment infrastructure that worked for humans at checkout was not designed for agents acting autonomously.
This article covers what agentic commerce is and how the new infrastructure actually moves money on behalf of an autonomous agent.
What is agentic commerce?
Agentic commerce is a transaction model in which an AI agent, not a human, initiates and completes the purchase. The user defines the rules upfront. The agent operates within those rules without further intervention.
It sits at the intersection of three existing categories: e-commerce (the merchant side), AI agents (the actor), and digital identity and authentication (the trust layer).
How agentic commerce differs from traditional e-commerce
Traditional e-commerce assumes a human at the keyboard. The flow is: search, evaluate, click, authenticate, pay. Every step requires human attention.
Agentic commerce compresses this to: instruct, delegate, observe. The user issues a single instruction ("pay vendor invoices up to $5,000 each"), and the agent does the rest.
The key difference is not speed. It is that the agent acts as the buyer, not as a tool used by the buyer.
How agentic AI differs from predictive and generative AI
Predictive AI forecasts. Generative AI creates content. Agentic AI takes action. It can plan, decide, execute, and adapt across multiple steps. In commerce, that means moving beyond product recommendations to completed transactions.
Why agentic commerce is happening now
How did we get to agents autonomously making purchases for us?
- Consumer behaviour shifted toward AI-assisted shopping, with AI-driven referral traffic to US retail sites growing 393% year over year in Q1 2026.
- Card networks and platforms responded with agent-specific frameworks: Visa launched Visa Intelligent Commerce, Mastercard launched Agent Pay in April 2025, Stripe and OpenAI co-launched the Agentic Commerce Protocol (ACP) in September 2025, and Coinbase shipped x402, an HTTP-native payment protocol for machine-to-machine transactions. These standards are mostly complementary, each addressing a different layer.
And the economic case is sizing up. McKinsey forecasts between $3 trillion and $5 trillion in agentic commerce sales globally by 2030.
How agentic AI executes payments in real time
When an AI agent acts on behalf of a user, a payment is no longer a single human action. It becomes a sequence of three things happening in real time: the agent notices that a purchase is needed, the transaction is evaluated against the user's rules, and the payment is routed through to settlement. Each step is automated; each step is governed by the policy the user set upfront.
1. Event detection and context gathering
The agent first has to recognise that a payment is needed. The trigger might be an incoming invoice, a usage threshold being crossed, a calendar event, or an explicit instruction from the user. The agent then pulls the context that justifies the purchase: who the merchant is, what the amount should be, whether it fits the user's pre-set policy, what the optimal timing or terms look like. None of this requires a person at the keyboard.
2. Policy evaluation and decision logic
With the event and context in hand, the transaction is evaluated against the user's rules. Is the merchant approved? Is the amount within the cap? Is the agent itself authorised for this category? Frameworks like VIC, Agent Pay, and ACP each implement this evaluation differently, but the principle is the same: the rules are enforced at a layer the agent cannot bypass, and a transaction that fails the check is declined before it ever reaches settlement.
3. Routing and settlement
If the transaction passes evaluation, the agent presents a scoped credential at the merchant's checkout. The payment network authorises the transaction in real time and routes it for settlement. To the merchant, the payment looks ordinary. To the user, the moment they signed the original instruction was the last time their attention was required.
Business use cases for agentic payments
The mechanics translate into a small set of repeatable scenarios platforms are starting to ship.
Automated vendor and supplier payments
An AI agent monitors invoices, validates them against pre-set spending rules and approved vendors, and completes payment without manual approval. Spend stays inside the user-defined scope (categories, limits, validity window). The finance team reviews exceptions, not every transaction.
Subscription and recurring payments
Software renewals, recurring vendor charges, and usage-based billing are the natural fit for agent-led flows. The agent tracks usage against tier limits, flags overages, and handles renewals within the signed policy. No manual re-entry of card details each cycle, no missed renewals, no over-spend in the cracks.
Cross-border B2B payment automation
Cross-border supplier payments are slow because they were designed for batched manual review. An agent can route payments across markets, select the optimal currency and timing, and complete settlement, while the signed policy keeps everything inside an approved scope.
Agent-assisted procurement
An AI agent receives a purchase request, validates it against pre-set spending policy, selects the right vendor, and completes payment, all before a human reviews the receipt.
How Reap does Agentic Commerce
Reap operates at the card-issuing layer underneath agent platforms. As one of the first licensed issuers to bring Visa Intelligent Commerce live, Reap agentic payment provides agent-specific tokenized credentials, programmable spend controls, and passkey-based authentication through a single API. PCI certification, KYC/KYB pipelines, fraud operations, and network integration sit at our layer.
If you are an AI platform, fintech, or developer looking to become an Agent Enabler shipping agent-initiated checkout, embedded purchase flows, or autonomous procurement, get in touch.